A CHASM of Mediocrity - Procedurally Generated Metroidvanias

Two things: First: The game at 1:01 is A Robot Named Fight. It's a "Roguevania". Second thing: I have a video about "Roguevanias", comparing Dead Cells and A Robot Named Fight on this channel. Check it out! ► https://youtu.be/LpceV7GD6fg

You missed another key part of that red tower/ice beam sequence: as you leave the ice beam room, you have to use the ice beam on the same enemy type as those that patrol the red tower. This immediately brings the red tower back into your memory and the correct way to progress is revealed. It's genius.

My main problem with Chasm is too many of the rooms are hallways with 1v1 fights one after another. I was hoping to see more variety where you'd have to deal with multiple threats at the same time.

Those are some damn-nice 16bit style graphics though.

Just started playing Hollow Knight. That glimpse at the full map. Holy. Crap.

"You find an item that lets Chasm crawl" Gold.

To be fair, the castlevania "animation skip when you land" thing is very common in all metroidvania, and something I actually expect them to do. And for good reason, gameplay tends to get very stiff when you can't do that

tldr the beauty of metroidvania is that it is a linear game, in a non linear world, the progress is pre planned to be good

It's all about how the procedural generation is implemented. You need to be creative about how it's going to work in order to get something that feels cohesive and is fun to play. Mark Brown has a good video analyzing the methods used in Spelunky, which worked well for that game. Doing procedural generation right is hard as balls, much more work than just building the levels yourself, but if you can pull it off you can end up with some amazing games.

Salt and Sanctuary is a good example of a sprawling map that's complex, and La-Mulana is a good example of a tight map that feels large because of its complexity.

So, Chasm could have been a good metroidvania but instead is a machine that "creates" a different mediocre metroidvania every time you start a game.

See, I love almost everything about Chasm. the fact that I know my map is not the same as someone else's map simply fascinates me. While playing, I seem to have been given a very good seed as most enemy types are spread out, and secrets are evenly dispersed, forcing me to backtrack to finally complete the map(Gosh that's so satisfying.) Not really having played many Castlevania games, I didn't see the copied elements except the spells, which are a total joy to use in Chasm, unlike all Castlevania games I've played. Which the ones I played made me choose which spell I liked most, and gave me a sense of paranoia about seeing another spell on the ground, then proceeding to dodge around it, often times getting hit by the everything else in the area. In terms of hidden connections, going back to my great seed I was given, the secret path from the Gardens to the Catacombs put me directly near a warp point, which was amazing, as the closest one was way too far away to be used effectively. All in all, the game is successful in my opinion. And that's really what makes these discussions fun, hearing someone else's point of view.

Castlevania LoS: Mirror of Fate was not a particularly good game, but the ability to take notes and stick them at any point on the map was a simple yet BRILLIANT addition that should have become a new standard for Metroidvanias. I give Hollow Knight a pass because it handled maps in a unique way in general, but there is no goddamn reason for ANY other Metroidvania to lack this note-sticking feature anymore. Metroidvanias are my favorite 2D genre bar none, but needing to remember the location of every instance of an inaccessible item or area with an obvious solution (i.e. I need to come back when I can crawl, when I can double jump, etc) is an archaic trope that should not be artificially forced on players anymore.

Making a good procedurally generated Metroidvania is possible, but extremely difficult and effort-intensive - you'd need to design a TON of rooms, including rooms and versions of bosses that can be handled with entirely different combinations of power-ups, and enough memorable setpieces and bosses to prevent too much repetition (at LEAST five times as many as one run will ever contain). That's after setting up the generation engine to determine what the power-up sequence is for this iteration of the gameworld, and then which zones get those power-ups, and then how the zones interconnect. After all of this, the engine must also sort these numerous zones in such a fashion that the overall map is coherent, then arrange the rooms within the zones to make those coherent and interesting as well. All told, it'd be easier and more profitable to just design four or five whole game worlds and sell each of them as its own title.

Rather than spending all that time developing procedural generation so Chasm could have infinite crappy maps, what if they had hand-crafted more maps than people can generally remember accurately - say, a dozen - and picked a map at random every time you start a game? Might be the best of both worlds. Even if none of the maps are as superb as Super Metroid's, they'd surely be better than whatever is vomited out procedurally, and still offer far more replayability than 99% of gamers could ever ask for.

This might be tangential, but there is one massive problem with proceduraly generated content that is often overlooked - your brain. It is a fantastic pattern recognition to a point it can recognize a pattern even when it is faint and quite honestly was a mere accident of random chance. And it is also a fantastic generalization and abstraction machine. For example, No man's sky. Basic complaint I heard was "it is different, but really, it is mostly just the same samenesss ". And quite honestly, it is what I feel about Deadcells after the few first hours of play. The thing is, when it is used to generate content that your brain will automatically generalize and abstract, so that only the general pattern (rather than the technically specific details), you spent a whole bunch of time accomplishing absolutely nothing. Actually, less than nothing. With a fixed layout, you can get the player to appreciate the attention to detail in layouting the place and designing at least in some level of consciousness. The layout is entirely random? You lose that and on top of that the player is generalizing your content as "the sewer looking level" and the "the sewer looking level, but in yellow", which isn't a positive trade-off. L4D is still imo one of the best examples of procedural generation. It is reserved exactly to the sort of thing the brain wouldn't immediately generalize and abstract the entirety of it. What enemy comes up next, in which way, when, which weapons you find, etc. Those are things you can't afford to "abstract and generalize" when they're random. Scenery? Simple layout? "Different" species that act in just as many finitely many ways so it is "the same shit as the other cat, but this one is purple with spots and weird tail"? All this stuff that, other than the first couple of minutes/hours observing the visuals will have no real impact on how you play the game? Easy af to become oblivious to the details (aka irrelevant differences).

I think it's quite telling that dungeon crawling for special loot items, a feature that was included in the original version of Minecraft, was completely ignored by the fanbase in favor of the open exploration and crafting. Procedural generation can create a lot of content to play, but it can't make the result feel rewarding. For a Metroidvania this is an even bigger problem; randomness clashes horribly with an upgrade-based exploration system. You need structure to have a good game of this kind. By the way, people hugely overestimate the amount of backtracking classic Metroidvanias had. I just replayed Metroid Prime, and it's surprising how neatly that game is designed to almost always permit you to reach the next upgrade quickly even if you've left the critical path. Super Metroid was also much more linear than people remember. Long sections of backtracking in both of these games generally means you're fooling around, not that the game is wasting your time. ...Also by the way, was that a SNES Classic controller you had around your neck at the beginning?

I just don't understand the allure of procedurally generated metroidvanias. In SotN, everything was always the same; I knew exactly where the library was, the coliseum, the catacombs, and it's still a wonderful game. I never wished it was randomized. The graphics and soundtrack were superb and I can play the game today and have a great time. Maybe I'm just boomin' here.

Wow, can't believe NOBODY said it. Fine, I'll say it, then: ROGUE LEGACY. Enemies designed for particular environments combined with procedurally generated maps. If you make the enemies and environments a little more cohesive, you can certainly re-combobulate the Rogue Legacy elements into something very much like Chasm. EDIT: I didn't say Rogue Legacy IS a metroidvania, I said it has almost all the parts. I'll expand on my last sentence. Scatter Rogue Legacy's classes/equipment/runes in the castle as items, put up some gates that limit exploration and force the finding of certain abilities before exploring further, and you're headed in a decidedly more Metroidvania direction, except unlike Chasm, you already have area-specific monsters and environmental hazards. Almost all of Chasm's gimmicks, but arguably better executed with less "saminess" of hallways and random enemies plopped haphazardly all over the map.

It seems like this game could do with a dependency tree, they probably have one but it seems kinda "off." A dependency tree that would label certain rooms and pathways a "dependent" on a power up, or skill. Once you have that create two separate layout generators, one that crafts a linear path, and one that creates an open space. These two systems would alternate, and randomly place items which would then inform the generators what can be done. Now create a 3rd generator which would essentially craft a linear path backwards which would use the dependency tree to add one way obstacles and then an exit at which point one of the two generators would takeover. Now I have no Idea if this would create a more interesting layout than what CHASM offers, it's just my speculation that this would create a system that would be playable (and unique, at least for the first 2 or 3 playthroughs) I would also incorporate secondary items which would give similar powers, but these would be hidden by the generators, and would not influence the generators. These items, if found, could get you through future obstacles, and in theory could give you a puzzle in which you have successfully collected item A early, which let you bypass a section, but in that section held items B and C which you'll need later on.